search

pvlib / pvlib-python

A set of documented functions for simulating the performance of photovoltaic energy systems.
https://pvlib-python.readthedocs.io
BSD 3-Clause "New" or "Revised" License
1.19k stars 1k forks source link

Adding New Single Diode Model For CdTe #163

Closed crimmj closed 6 years ago

crimmj commented 8 years ago

Would like to add a new diode model(s) appropriate to non c-Si modules. There are some additional parameters in the model(s). Any thoughts on the correct structured way to do this? Would this function or functions be added to pvsystem.py? Do modules need to be a class if they have different appropriate performance models?

wholmgren commented 8 years ago

Sounds useful to me. The existing functions that use module parameters will work with anything that supports dictionary-like lookups. Most of the time, people will be passing in Series objects from the SAM DataFrames, but plain dictionaries should work too. I would say that we should keep things simple and use a similar interface for new functions.

cwhanse commented 6 years ago

This request essentially asks to implement the PVsyst model for a-Si, which includes a new term to account for recombination loss in the intrinsic layer (see http://files.pvsyst.com/help/thinfilm_recombinationloss.htm

PVsyst provides a default value for the single new parameter (di^2/mu teff = 1.4V).

Implementing this new model depends on having calcparams_PVsyst see #470 and will require a new function to solve the diode equation with the recombination current term.

mikofski commented 6 years ago

I have this model already, I can try to add this

cwhanse commented 6 years ago

@mikofski happy to have a volunteer! Are you thinking to implement the PVsyst model (addressing both #470 and #163) or only the variation of singlediode to solve the current-voltage equation? If you need help with #470 I can do the calcparams_pvsyst component.

mikofski commented 6 years ago

If the recombination loss current is already included in the MATLAB version, then I guess we should just port that over here. @cwhanse are you planning to do that or are you looking for volunteers?

cwhanse commented 6 years ago

I have some MATLAB code which isn't in PVLib. I didn't find an elegant solution, it's brute force getting the current at each voltage.

mikofski commented 6 years ago

Okay let's collaborate then. I use the same explicit approach for PVSyst as I do for any SDM. Then either interpolate or Newton method or bisection for specific points or the max p. I'll try to put something up this week. I think this counts as work don't you? On a related note, I added an issue on SAM SSC on the recombination current for thin film. I think it should be added to the CEC and DeSoto models, the J. Merten, Mermoud, and Lejeune seem to indicate that an error would be present with thin film for any SDM near Voc missing this term. Do you have an opinion? Maybe we can collaborate with FirstSolar.

cwhanse commented 6 years ago

I think I need to be clear on the plan here. This issue asks for a new function (counterpart to pvsystem.singlediode that solves

I = IL - IL x mu/[Vbi - (V+I x Rs)] - I0 x [exp((V+I x Rs)/(nNsVth))-1] - (V + I x Rs)/Rsh

returning the IV curve. The second term on the right is the recombination current.

470 asks for calcparams_PVsyst which is related, but basically independent, because the parameters unique to the recombination term (mu, Vbi) are constants.

We can work together to implement singlediode_recomb or whatever name we settle on. I'll email my Matlab function to you.

mikofski commented 6 years ago

Yes, I'm clear. What I was asking about was that since the only difference between the existing single diode model (SDM) and a model that included the thin film recombination loss current is the term:

I_rec = I_L * d2_mu_tau / (V_bi - V_diode)

then:

As I said previously, the solution for either is exactly the same: specify a bunch of diode voltages, solve for cell current then back calculate the cell voltage. For specific voltages and max power, then you can either interpolate, use Newton, or use a bisection search method. (Maybe Lambert-W also works but I'm not as familiar with that formulation?)

I understand that calcparams_pvsyst merely generates the 5 parameters that SDM uses since PVSyst has different auxiliary equations, so I understand that's the focus of #470.

So, as I asked above, for #163 we could either implement a new SDM method for thin film with recombination losses or just combine it with the existing SDM.

Regarding:

I think it should be added to the CEC and DeSoto models, the J. Merten, Mermoud, and Lejeune seem to indicate that an error would be present with thin film for any SDM near Voc missing this term. Do you have an opinion? Maybe we can collaborate with FirstSolar.

Sorry to be so vague with that comment. What I meant was the question I asked above: shouldn't we just include the recombination loss current into the existing pvsystem.singlediode such that it would be available to any calcparam_*() method, not just PVSyst? I was thinking that by collaborating with FirstSolar we could determine whether or not the existing SDM without the recombination loss current exhibits the errors that J. Merten, Mermoud, and Lejeune observed.

adriesse commented 6 years ago

I think it makes sense to incorporate this feature into the SDM functions. The various models can then use the feature as needed. Maybe it even opens up some unexplored territory.

cwhanse commented 6 years ago

I'm tracking with you now @mikofski If we use the diode voltage method to solve the IV curve this new term is not a challenge. But singlediode and its helpers, i_from_v and v_from_i are using the LambertW method which doesn't extend when the recombination current is added to the single diode equation.

We had a long conversation about changing to the diode voltage approach #409 which hasn't been merged. We need to resolve that before moving ahead with this issue.

Since it has been a while, @mikofski can you remind us what issues remain with #409?

mikofski commented 6 years ago

I've started trying to update #409, there's a list of TODO[s] that @wholmgren left for me to check off like fixing docstrings, better names (newton vs. "fast", etc.) and moving vendorized scipy.optimized.newton to _tools since it won't be merged until November (if ever ☹️ )

mikofski commented 6 years ago

TODO: see this comment from #409 to add Irecomb to bishop88

Yes. Set d2mutau=0.0 and Vbi=0.9, user can change the values with kwargs if wanted.

mikofski commented 6 years ago

TL;DR:

Should we always limit IV curves for CdTe (and a:Si?) to the 1st quadrant only, or to a voltage in the 4th quadrant that is very close to open circuit voltage but much less than the intrinsic voltage?

More Discussion:

@cwhanse as voltage approaches the builtin voltage, the expression for the recombination loss current asymptotes to infinity:

diode_voltage = voltage + current * series_resistance
Irecombination = photocurrent * d2mutau / (intrinsic_voltage - diode_voltage)

@jdnewmil has suggested gradually turning off the recombination somewhere between open circuit (V_oc) and the intrinsic voltage of the thin layer, which should always be greater than V_oc.

There is nothing in Meromoud's Performance Assessment ... that suggests what to do.

Do you have any suggestions?

First to be clear, the recombination loss error, does not affect all thin films. For example, CIS/CIGS are not affected. Merten et al and Mermoud et al only observed the recombination loss for amorphous silicon (a:Si) and cadmium telluride (CdTe).

My idea would be to just limit IV curves for CdTe (and a:Si?) to the 1st quadrant only. This avoids the problem. There's no reason to go into the 4th quadrant for thin film because they typically don't have bypass diodes and therefore will fail badly if there is any partial non-uniform shade, ie: only row-to-row shade that effects the entire array (except the front row) evenly. Also most CdTe (and a:Si?) are typically only installed on trackers, and are only shading when the sun is low in the sky since most CdTe (and a:Si?) do not backtrack. Since the trackers are aligned N-S, the sun is nearly due east or west during sunrise/sunset, and the irradiance is very low. String voltage mismatch might be an issue at sunrise/sunset because of edge effects since the sun is not exactly due east or west, but the voltages should all be well below the intrinsic voltage due to the low irradiance, and voltage mismatch would be small because the sun is nearly due east or west.

Thanks for your comments.

cwhanse commented 6 years ago

Yes, 1st quadrant only is fine from my point of view.

jdnewmil commented 6 years ago

Really depends on what you plan to do with it. If you plan to study voltage mismatch then individual curves can operate in the fourth quadrant even while the overall array may not. My conclusion was that doing one quadrant only was unacceptable for my purposes.

On July 11, 2018 10:22:04 AM PDT, Cliff Hansen notifications@github.com wrote:

Yes, 1st quadrant only is fine from my point of view.

-- Sent from my phone. Please excuse my brevity.

cwhanse commented 6 years ago

@jdnewmil your use case is of interest to me also. The model we are implementing is chosen because its being used in PVsyst. I don't know if PVsyst does calculations outside the 1st quadrant.

I worry about applying this model at high external voltage. The way I read the reference is that this particular model for the recombination current loss derives from some assumptions that are valid at low external voltage, but questionable as the system moves beyond Voc. I'm not eager to spend effort to make this particular calculation reliable as diode voltage approaches the intrinsic voltage parameter value. I think we'd be better served looking for a modeling approach which could be justified at these voltages, e.g., Augusto et al (207) Analysis of the recombination mechanisms of a silicon solar cell with low bandgap-voltage offset.

mikofski commented 6 years ago

Proposed changes as discussed in #409

What are good names for the new arguments?

remaining questions:

I'll submit a PR soon, but it'd be nice to get some feedback on suggested names, any mistakes in my derivatives, or ideas about arguments and code paradigms. Thanks!

cwhanse commented 6 years ago

voltage_builtin I think . I prefer to put the quantity first. The Mertens paper references earlier work that uses the term 'built-in voltage' and that term seems to have stuck. 'intrinsic voltage' appears to be a PVsyst term. Users should be permitted to set voltage_builtin via a kwarg but we should supply 0.9V as the default.

We could consider adding a warning that this model is only considered appropriate for amorphous silicon cells. Validity at high forward voltage is a different issue...

mikofski commented 6 years ago

... and CdTe, but not CIGS - Yes I agree, this should be emphasized. What to call d2mutau ? My vote is to leave it as is.

jdnewmil commented 6 years ago

It is a composite of variables, so yes, leave it as-is.

On July 13, 2018 1:43:59 PM PDT, Mark Mikofski notifications@github.com wrote:

... and CdTe, but not CIGS - Yes I agree, this should be emphasized. What to call d2mutau ? My vote is to leave it as is.

-- Sent from my phone. Please excuse my brevity.

adriesse commented 5 years ago

Not sure it is pythonic (or gitic) to comment on a closed issues, but it seems it would be really useful to have d2mutau and NsVbi accepted in bishop88_mpp, or something like that. Has anyone got that on their radar screens?

cwhanse commented 5 years ago

Not on my radar. I agree it would be useful. Let's open a new issue for the enhancement request.

mikofski commented 5 years ago

OK, but bishop88 does already take these arguments: https://github.com/pvlib/pvlib-python/blob/0a14b273517a082603cc157faa88a5ab0ac4cac9/pvlib/singlediode.py#L70-L72

Seems like an oversight that they're not in bishop88_mpp, bishop88_i_from_v, or bishop88_i_from_v. I believe this may be an easy fix?

adriesse commented 5 years ago

Seemed like something that fell between the cracks. Is it really just a matter of passing the parameters through?

mikofski commented 5 years ago

yes

adriesse commented 5 years ago

@mikofski Care to look in my bishop branch? It's all I had time for today; I think it works. Maybe a PR next week.